FFT suspension analysis

Here are our CFD links and discussions about aerodynamics, suspension, driver safety and tyres. Please stick to F1 on this forum.

Post Thu May 10, 2012 1:25 pm

Well do they not match reality?
I must say that I am surprised by the shape of the curve and thought that it is strange but I have also seen curves from rfactor which seem to make sense. I think it is just a question of clicking on the correct buttons in motec to setup the axes in the right way.

If it does have a influence on laptimes is a totally different question. I say it probably does not bring you a laptime improvement but that does not mean that it is useless to spent time with it because it is still possible to learn from it and draw conclusions for a real race car.
mep
 
Joined: 11 Oct 2003
Location: Germany

Post Thu May 10, 2012 2:40 pm

mep wrote:Well do they not match reality?


Why would they? What validation do we have of the commercially available video game against the real world? None. Can't say one way or the other.

Moreover is the point of the SIGNIFICANCE of it, and if that is captured appropriately in rFactor - including the tire model. Would be dangerous to draw the wrong conclusions from tuning things in the game that might not have anything to do with the real response.
Grip is a four letter word.

2 is the new #1.
Jersey Tom
 
Joined: 29 May 2006
Location: Huntersville, NC

Post Thu May 10, 2012 7:41 pm

Once you go above about 5Hz, you are getting into the range where even professional simulation tools costing several 10s of thousands of dollars are not even accurate.

Sure, if you make a change on the car you will see a change in the FFT, but there is no way this will have anything to do with reality.

Tim
Not an engineer at Caterham F1
Tim.Wright
 
Joined: 13 Feb 2009

Post Sat May 12, 2012 12:18 am

JT, as usual, we are not to discuss if the method is wrong or not. Joncho made the questions to learn how to interpret the curves, not if they can be translated from a sim to reality.

Joncho, you have got PM.
"You need great passion, because everything you do with great pleasure, you do well." -Juan Manuel Fangio

"I have no idols. I admire work, dedication and competence." -Ayrton Senna
Belatti
 
Joined: 10 Jul 2007
Location: Argentina

Post Sat May 12, 2012 12:42 am

joncho wrote:As far as I read in other parts of the forum, the post rig tests is not with constant amplitude, it vary the amplitude depending of the frequency, so that would an inconvenience for my rfactor test, and also is that I dont know in depth the ecuations of the simulator and so it makes me doubt.


The underlined. Anyone knows how does the RF does this? I have been looking a while ago to the files where the mods are defined and there are a lot of aspects of the racing cars that are stored (chasis and tyres properties). It looks like probably the model may be not that far away than a costly chasis sim.
"You need great passion, because everything you do with great pleasure, you do well." -Juan Manuel Fangio

"I have no idols. I admire work, dedication and competence." -Ayrton Senna
Belatti
 
Joined: 10 Jul 2007
Location: Argentina

Post Thu May 24, 2012 5:33 am

I slowly start to understand the meaning of the fft graph when I playing with scale (log or not log) and changing the testing surface (asphalt) and car characteristics like spring and dampers and tyres.

I found many files really interesting like this one
http://enpub.fulton.asu.edu/jalali/EEE598_Fall07/fundamentals_FFT_based_analysis.pdf

The problem to see the correct data is called windows leakage.

This graph explain all to me

Image
joncho
 
Joined: 9 May 2012

Post Thu May 24, 2012 5:40 am

Jersey Tom wrote:
mep wrote:Well do they not match reality?

Why would they? What validation do we have of the commercially available video game against the real world? None. Can't say one way or the other.


I compromise here to spend time to achieve this. A model in matlab (with Belatti help :D ) with all rfactor numbers like spring and damper of the car and the tires and FFT graphic

vs.

The FFT of motec

The validation is ok? It will take time because I have to learn many things before work on
joncho
 
Joined: 9 May 2012

Post Mon Apr 29, 2013 1:53 am

Hi guys,

I am doing a final year project for a final year engineering assignment correlating real data of a GP2 car with a model I produced using ChassisSim car simulation software.

I am trying to get an FFT plot working in MATLAB but have had issues with the spectral density numbers. Is there an easy way to normalise these values to an amplitude?

Any help would be much appreciated.

Regards

Ed
fasteddiev
 
Joined: 29 Apr 2013

Post Mon Apr 29, 2013 2:08 am

...not quite sure what you mean, to be honest. Then again my fundamental knowledge of some of this stuff isn't as great as my ability to hack stuff together.

Could you be more specific of what issues you're having?
Grip is a four letter word.

2 is the new #1.
Jersey Tom
 
Joined: 29 May 2006
Location: Huntersville, NC

Post Mon Apr 29, 2013 2:16 am

So I am producing FFT plots in MATLAB using the pwelch method.

It is giving me OK frequency results but not enough resolution, and my Y-axis is the power spectral density. I would like to have this as a smaller magnitude us I can normalise results between my 'real' data at 100hz sample rate and my simulated results at 50hz sampling rate.

x = pwelch ( displacment, sample length, 0, sample length, sample rate)

That is the formula I am using, I hope that makes sense.

Thanks for the help

Ed
fasteddiev
 
Joined: 29 Apr 2013

Post Mon Apr 29, 2013 2:31 am

There is a 'fft' function in MATLAB... rather than having to use 'pwelch' to compute the PSD.

Are you trying to compare sim data analysis in MATLAB and "real" data analysis in a different software package, or what? I would think the most apples-to-apples comparison would be to bring both data sets into MATLAB, downsample your track data to 50 Hz, and then use the same analysis method (be it 'pwelch' or 'fft' or other).
Grip is a four letter word.

2 is the new #1.
Jersey Tom
 
Joined: 29 May 2006
Location: Huntersville, NC

Post Mon Apr 29, 2013 4:18 am

The matlab command for downsampling is decim

i would be very wary about comparing psds from different software, as Tom said, import both the track and model runs as time histories and analyse them in exactly the same fashion.

I use fft not pwelch because I wanted exact traceability of what was going on, and while I can check fft I would have to trust pwelch.

You'll find that establishing the vertical scaling of a psd is not exactly straightforward, but the help does give you a complete description. I synthesised a test signal to check that the scaling was right.

If you want higher res with an fft just double the number of samples in a frame. You'll need a new scaling factor!
Greg Locock
 
Joined: 29 Jun 2012

Post Mon Apr 29, 2013 11:00 am

fasteddiev wrote:So I am producing FFT plots in MATLAB using the pwelch method.

Without disagreeing with either JT or Greg, & admitting I have never used pwelch, a quick look suggests that is the kind of thing I would expect to have to do to obtain good frequency domain estimates*. Scaling issues can be overcome by computing frequency response functions, assuming that you have both an "input" & an "output" and you can replicate the same process with your model results.

fasteddiev wrote:It is giving me OK frequency results but not enough resolution...

Frequency_interval = acquisition_rate/N. Hence you can increase N, as Greg suggests, or decimate the data (reducing the acquisition rate). However, I expect you really need more data, since the variance error is proportional to the number of independent transforms, & increasing N or decimating a fixed length of data simply increases the variance error.

* It should be noted that an FFT calculation is actually a "Discrete Fourier Series" estimator, the implication being that the original data series is assumed to be exactly periodic. If that is not case, then the result will be contaminated by the end condition (the discontinuities between the end and beginning of the data sample). "Hanning" or other similar weighting functions are designed to eliminate any discontinuities, but they also eliminate part of the data sequence, which can be recovered (to an extent) by overlapping successive transforms - assuming that the process computes the average of multiple transforms.

The average of multiple transforms is computed for exactly the same reason that averages are usually taken of any measurements (to improve reliability).

The result is a transform whose scaling depends on the data, the "window" used and the precise nature of the overlapping algorithm. The scaling can be determined experimentally (as suggested by Greg), or cancelled out by computing a Frequency Response Function.
DaveW
 
Joined: 14 Apr 2009

Post Mon Apr 29, 2013 4:01 pm

Thanks guys, so have down-sampled my 100hz model and have had these results.

I was looking to get correlation between the body and hub modes between real and simulated data and I think there is fairly good correlation, given the bump profiling is produced within the simulation software. I am trying to rectify the body mode excitation levels but the hub mode looks correct for the car.

Image

Image

The damper histograms look reasonable as well, apart from the low speed on the front.

Cheers

Ed
fasteddiev
 
Joined: 29 Apr 2013

Post Mon Apr 29, 2013 5:05 pm

fasteddiev wrote:I was looking to get correlation between the body and hub modes between real and simulated data and I think there is fairly good correlation, given the bump profiling is produced within the simulation software. I am trying to rectify the body mode excitation levels but the hub mode looks correct for the car.

Would appreciate more information about your plots. e.g. What are the "magnitudes" shown (position, load, acceleration?), and where was the data gathered (track tests or rig tests). Do you know the natural frequencies of your model? Plotting on a white background might be helpful, to me at any rate.
DaveW
 
Joined: 14 Apr 2009

PreviousNext

Return to Aerodynamics, chassis and tyres

Who is online

Users browsing this forum: Exabot [Bot] and 5 guests